Goldman estimates central banks bought 79 tonnes in August. The reported total was 39. The Dutch spent half a year moving 86 tonnes toward London, and most of that move was a sale and a purchase, not a ship.
This is not a recommendation to buy or sell gold, a gold fund, a mining share, or any other security. It is not a price target. Gold can fall while a central bank is still buying. A vault decision is not a bid for your stock.
Here is the only idea. The price is in a fight with the Federal Reserve. The bars are in a fight with the vault. Those are different facts. An exchange-traded fund can be sold because a rate went up. A reserve manager who moves metal out of New York is not asking the rate for permission. Do not trade one of those facts as if it were the other. And do not let a round viral number stand in for the scale.
Two numbers for one month
On Monday, October 19, 2026, a Goldman Sachs note, as reported by Reuters, said central banks bought an estimated 79 tonnes of gold in August. The bank’s own guess for July had been 44 tonnes. The pre-2022 average in that note is 17 tonnes a month. Seventy-nine divided by 17 is about 4.6. That is more than four times the old pace. It is not five times. A blog that wrote “nearly five times” rounded toward the drama.
The same day, the number you can actually read off official reports is smaller. The World Gold Council, using the International Monetary Fund and the central banks themselves, said reported net buying in August was 39 tonnes. Year to date, the reported total was 170 tonnes. Thirty-nine is still more than double the old monthly average of 17. It is not 79.
Hold both. Seventy-nine is a model of what Goldman thinks was bought, including metal it believes did not show up in the official line. Thirty-nine is what the officials were willing to print. The gap is 40 tonnes. That gap is the argument. It is not, by itself, a licence to double China’s number and call the rest a secret.
China was the largest identifiable buyer in Goldman’s August count, at an estimated 25 tonnes. The People’s Bank of China reported an addition of 20.2 tonnes for August. The World Gold Council rounded that reported buy to 20 tonnes and called it the 22nd month in a row. China’s reported reserves stood near 2,387 tonnes, about 9 percent of its total reserves, after about 80 tonnes of reported buying in the first eight months of 2026.
Twenty-five versus 20.2 is a real gap. It is about five tonnes. It is not “twice.” A chart headline that says China bought twice what it admitted does not survive August’s arithmetic. Five tonnes can be a London over-the-counter estimate, a timing difference, or a bar the model thinks it saw. It is not a second China.
The other reported buyers in the Council’s August note were not a secret either. Poland bought 8 tonnes that month. Poland’s reported total for the year stood at 98 tonnes, and its stock at 648 tonnes, with a public target of 700. Uzbekistan was also on the list. The Council did not need a back channel to find those names. The quiet part of the month is whatever sits in the 40 tonnes between 39 and 79, and Goldman did not publish that remainder as a country table in the reports used here.
What the fast pace is, and what it is not
A reading of Goldman’s nowcast, in an account published the same evening, put the three-month seasonally adjusted pace at 85 tonnes a month. That account said the July nowcast had been 91, and a June update near 100. If those readings are right, the pace cooled. It cooled from something over five times the old average to something near five times. A slowdown from 100 to 85 is not a return to 17. Calling it a bust is a mood. Calling it unchanged is also wrong.
Multiply 85 by 12 and you get about 1,020 tonnes a year. The old 17 a month is about 200 tonnes a year. That is the napkin the viral note wants you to remember: official buying, annualized, about five times the pace from before Russia’s reserves were frozen in 2022. Goldman has said that freeze is why the bid changed, and that the change looks structural. Emerging-market banks, and China in particular, still look underweight gold against the allocations the bank thinks are plausible. The motives it names are dull and large. Less reliance on G7 reserve currencies. Geopolitics. Financial risk.
Now put the napkin next to the printed book. One hundred and seventy reported tonnes in eight months is a straight-line pace near 255 tonnes a year. That is above the old 200. It is not 1,000. The thousand-tonne year is the model, stretched out for twelve months, on a pace that has already ticked down from the June reading. A model can be the better description of the world. It is still a model. An investor who books 1,000 tonnes as if the Council had weighed them has booked a forecast.
Goldman has already shown how it uses the model in a price. In a September 18 note it lifted its central-bank demand assumption to an average of 60 tonnes a month through 2026 and 2027. That is below the 85 it was counting, and far above 17. The bank said that official buying was nearly all of the reason it still looked for about a 23 percent rise in the gold price by the end of 2027, to $5,400 an ounce. It had cut its end-2026 fair value to $4,650. The long target stayed. The near target came down. That is a bank saying the floor is official demand and the ceiling, for now, is the Federal Reserve.
The vault is a different document
The buying tally is an argument about tonnes. The custody tally is an argument about address. They can move together. They do not have to.
In early September the Dutch central bank, De Nederlandsche Bank, said it had spent March through August relocating gold. The Associated Press report is the clean one. Some 86 metric tonnes were shifted toward London out of the pool that had sat in New York and Ottawa, a pool of about 313 tonnes. Before the move, New York held 31.3 percent of the Dutch stock and Ottawa 19.7 percent. After it, each city held 18.5 percent. The bank’s total stock did not change. It owns 612.4 tonnes. At the end of 2025 that stock was valued at 72.2 billion euros.
Read the method, because the method is the fact the viral note smeared. More than 27 tonnes were physically moved from the United States and Canada to the Dutch vault at Zeist. A similar amount then moved from Zeist to London. The rest, about 59 tonnes, was sold in New York, and the money was used to buy gold in London. Most of the “departure” from the New York Fed was a trade. The bar in Manhattan did not have to cross an ocean. A different bar, in London, took its place on the Dutch list.
That is still a decision about New York. It is not a thriller about a nighttime flight. Governor Olaf Sleijpen said the relocation improved the tradability of the reserves. He said the bank expects it will never need to use them, and that it still has to strengthen resilience and preparedness. The bank’s own line on the Bank of England is plainer than any blog. Gold there meets modern trade standards, is the easiest gold in the world to trade, and would be the most ready in a crisis. Gold in New York and Ottawa, it said, cannot be used as quickly or as directly.
Sit with that sentence. A European central bank did not say it was dumping dollars, and it did not say it was selling gold. It said the gold it kept in the Federal Reserve’s basement was the gold it could not use fast. Then it sold a large piece of that gold in New York and bought a London piece it says it can use. Preparedness, in this case, is a logistics word. It is also a trust word. They chose to say both.
A later account added that France had removed all of its gold from the New York Fed between July 2025 and January 2026. This note does not have the Banque de France tonne count in a primary release, so it will not invent one. The Dutch file is enough. One central bank wrote down the percentages, the dates, and the split between a ship and a sale. That file is the custody story you can audit. A chart that says “hundreds of tonnes went east” is a picture of the same worry. It is not a substitute for the Dutch footnotes.
What the charts are, and what they are not
The evening account of Goldman’s note read two custody lines off a chart. Foreign gold at the New York Fed down 40 tonnes in August. Custodial holdings at the Bank of England up 46 tonnes. It said July had already shown a 63-tonne rise at the Bank of England. Its rough read of the longer picture was about 200 tonnes gone from New York custody since early 2025, and about 500 tonnes added in London.
Treat those as a reading of a chart in a bank note, not as a customs form. Forty and 46 can be revised. Two hundred and five hundred are the words “around” and “roughly” doing their job. The Dutch 86 is not in that category. It is a named bank, a named window of months, and a named method.
There is a useful trap in the August pair, and the note, as that account renders it, already flags the trap. Goldman’s buying nowcast is built in the London over-the-counter market. It does not, on that account, pick up a change in whose name is on a bar already sitting at the Bank of England. If London custody rises by more than New York custody falls, some of the difference may be official buying the nowcast missed. In August the London rise was 46 and the New York drop was 40. The difference is 6 tonnes. That is narrow. It does not prove a hidden wave. It does prove you cannot add the custody chart to the 79 and pretend you have counted the world once.
A sale in New York and a buy in London, which is what the Dutch did with about 59 tonnes, will move both custody lines and may not be “new” gold at all. The world did not gain 59 tonnes because the Netherlands changed cities. The Netherlands gained a claim it says it can trade on a bad day. If you add that shift to Goldman’s 79, you will count a relocation as a purchase. The viral note leans on the gap. It does not always say which gap is a purchase and which gap is a change of address.
The price has its own boss
The same evening account put the spot price near $4,186 an ounce on Monday morning, citing Goldman’s materials desk. It said that was below the roughly $4,350 level at which the bank, last month, cut its year-end fair value to $4,650. It said the price was getting close to a $4,070 “near-term floor” the bank had drawn for a hawkish case.
The Federal Reserve has already hiked. Goldman’s economists, as of that September note, expected another hike in October. Higher rates have been a weight on exchange-traded fund demand. That is the part of the gold market that can leave without a meeting. A fund share is a promise. It is sold when the promise pays less than a Treasury bill, or when the holder is scared. A reserve manager who is underweight, and who is worried about where the bar sits, does not run that calculation the same way.
Goldman’s own split is the grown-up version of the viral headline. The Fed is the near-term drag. Central banks are the floor under the longer forecast. The floor is not a law. It is a flow assumption of 60 tonnes a month, inside a target of $5,400 that the bank has not earned just by repeating it. Between the spot price and that target sits a year and a half, another possible hike, and a model that says 79 in a month when the printed total says 39.
Not everyone at a bank desk buys the floor. The same account says Lawson Winder at Bank of America argued on October 12 that higher rates should keep a lid on gold. It says he preferred copper and uranium, and that he cut Kinross to underperform. It also says Société Générale has flagged demand destruction at these prices, with global demand in the first half of 2026 running under its ten-year average. Those are rival notes, not facts about the next print. They matter because a structural bid and a tired price can be true in the same week. August’s buyers did not stop the metal from sitting nearer $4,200 than $4,600.
What you can use, and what you cannot
You can use the gap between 79 and 39 as a question, not as a tonne. Each month, ask which number the next official print looks like. If reported buying stays near 40 while Goldman stays near 80, the model is claiming a shadow market twice the size of the press release. That claim has to keep coming true in some later, harder data, or it is a story about a model. If the reported number jumps toward the model, the shadow was real and merely late. China’s August gap, five tonnes, is too small to carry that shadow by itself. Someone else, or the model, has to.
You can use the Dutch file as a check on the word “buying.” A central bank can change the map of its gold without adding an ounce. When you see New York custody down and London custody up, ask how much was a sale and a purchase of equal size. That transfer still says something harsh about the New York vault. It does not say the world’s mines had to produce another 59 tonnes for the Dutch.
You can use Sleijpen’s sentence as the motive, because it is primary and because it is modest. He does not say the reserves will be spent. He says they must be ready, and that ready means London. A buyer who wants metal he can mobilize is a different buyer from a buyer who wants a line in a reserve report. The first buyer cares about the bar’s address. The second cares about the monthly tonne. Price is a third concern. He did not say he was waiting for a dip. He also did not say he would pay any price. Do not put those words in his mouth.
You cannot use any of this as a reason that a miner, a royalty, or a fund must rise. Kinross was cut to underperform in the account of Winder’s note for a rates reason, not a geology reason. A custody shift does not widen a mill. It does not cut a cost. It does not put gold in a deposit that was not there on Friday. If you own a miner, you own rocks, a permit, a cost, and a share count. The Dutch did not mine your rocks when they sold a bar in New York.
You also cannot use $5,400 as a schedule. It is one bank’s mark for the end of 2027, built mostly out of an assumed official bid of 60 tonnes a month. Spot, on the Monday reading, was near $4,186. The distance between those numbers is the bull case. The distance is not a debt the market owes you. The bank already cut the 2026 mark when the hike landed. It can cut the 2027 mark the same way.
What would make this reading wrong
This reading is wrong if the next official tables show August was much closer to 79 than to 39. Then the “reported” 39 was a lag, and the model was the scale. Early caution about a same-day note would still have been the right way to read a blog. It would be stale.
This reading is wrong if China’s later reserve file is revised up by tens of tonnes for August, and the five-tonne gap was only a clock. Then “they admit less than they buy” gets its month. It does not get it from 25 versus 20.2.
This reading is wrong if you hear “most of the Dutch move was a sale” as “nothing happened.” Fifty-nine tonnes sold in New York is a large exit from one custodian, even when a London bar replaces it. The percentage in New York fell from 31.3 to 18.5. That is a change of address on purpose. The purpose was written down. Crisis preparedness. Faster use. A vault that, in their words, cannot be used as quickly.
This reading is wrong if you treat the price sitting near $4,186, under a $4,650 year-end mark, as proof the buyers have left. Reported buying of 39 tonnes, and estimated buying of 79, both happened in a month when the price was already high enough for another bank to talk about demand destruction. High prices and official buying are not enemies in this cycle. They are roommates. Roommates still argue about the rent. The rent, this month, is the funds rate.
The idea, once
The gold price is arguing with the Federal Reserve. The bars are arguing with the vault. Goldman’s August estimate says the official bid is still more than four times the old monthly pace. The printed bid says it is more than double, not more than fourfold. China’s secret, on the numbers you can check, is about five tonnes, not a second China. The Dutch secret is not a secret. They sold about 59 tonnes in New York, shipped a smaller piece, and said London is where the gold can be used.
A fund can dump that argument into a hike. A reserve manager who is rewriting the address on the bar is doing a slower thing. He is deciding which basement he trusts when the phone rings. Until the next print tells you whether August was a 39-tonne month or a 79-tonne month, the only hard object on the table is that Dutch file. The price will keep shouting at the Fed. The file already picked a city.